Ska Imaging Achieves 14% Performance Boost with astroCAMP Co-Design Framework
Summarize this article with:
The impending operation of the Square Kilometre Array (SKA) presents a significant challenge, requiring sustainable processing of enormous datasets under strict power constraints. Denisa-Andreea Constantinescu, Rubén Rodríguez Álvarez, Jacques Morin, and colleagues address this challenge by introducing astroCAMP, a novel framework designed to guide the co-design of next-generation radio imaging pipelines and high-performance computing architectures. Current systems typically utilise only a small fraction of their potential, leading to inefficiency and high costs, and astroCAMP tackles this issue by providing a unified suite of metrics covering scientific fidelity, performance, sustainability, and economic factors. This work establishes standardised datasets and benchmarking tools, enabling rigorous evaluation of different hardware platforms, and demonstrates the potential for identifying optimal operating points for SKA-scale imaging, ultimately paving the way for more efficient and environmentally responsible astronomical research.
Scientists have developed astroCAMP, a novel framework designed to guide the co-design of next-generation radio imaging pipelines and high-performance computing architectures. Recognising that existing radio-interferometric pipelines typically achieve only 4-14% of hardware peak performance, scientists developed a comprehensive methodology to maximise scientific return within strict power constraints. This work introduces a unified metric suite encompassing scientific fidelity, computational performance, sustainability, and lifecycle economics, enabling a holistic assessment of imaging systems. To facilitate reproducible benchmarking, the team created standardised SKA-representative datasets and reference outputs, allowing for consistent comparison across CPUs, GPUs, and emerging accelerator technologies. These datasets, combined with a multi-objective co-design formulation, link scientific-quality constraints to critical metrics such as time-to-solution, energy-to-solution, and carbon-to-solution, all while considering total cost of ownership. The researchers employed the PREESM framework to demonstrate how astroCAMP enables multi-objective design-space exploration and the computation of Pareto fronts, identifying optimal trade-offs between competing objectives. A detailed analysis of existing imaging tools revealed significant gaps in evaluation practices, particularly regarding architecture-level metrics like power consumption and hardware utilisation. Notably, the study pioneers the inclusion of carbon-to-solution and carbon efficiency as first-class objectives, directly addressing the SKA’s sustainability goals and aligning with the UN Sustainable Development Goals. Current radio-interferometric pipelines typically achieve only 4 to 14% of peak hardware performance due to bottlenecks in memory and data input/output, resulting in inefficient energy use and high operational costs. This work introduces a comprehensive system for guiding the co-design of next-generation imaging pipelines and sustainable high-performance computing architectures. astroCAMP provides a unified metric suite covering scientific fidelity, computational performance, sustainability, and lifecycle economics, enabling researchers to evaluate trade-offs between these factors.
The team released standardised SKA-representative datasets and reference outputs, allowing for reproducible benchmarking across CPUs, GPUs, and emerging accelerator technologies. This facilitates rigorous testing and comparison of different hardware and software configurations under realistic SKA conditions. Experiments demonstrate that achieving the required efficiency for SKA-Low and SKA-Mid, which demands tens of GFLOP/s/W, is comparable to the performance of the most energy-efficient supercomputers currently available. However, existing imaging workloads are overwhelmingly limited by memory bandwidth and irregular data access patterns, hindering the realisation of peak performance.
The team evaluated co-design metrics using WSClean and IDG on an AMD EPYC 9334 processor and an NVIDIA H100 GPU, and further illustrated the use of astroCAMP for exploring design options with heterogeneous CPU-FPGA systems. This work identifies potential operating points for SKA-scale imaging deployments that balance performance, energy consumption, and cost. The research highlights that achieving the necessary efficiency requires a holistic approach encompassing algorithmic optimisation, domain-specific accelerators, and energy-aware orchestration. Researchers have developed a comprehensive system for guiding the co-design of both imaging pipelines and the underlying hardware architectures, explicitly targeting maximised scientific return within strict power and environmental constraints. This achievement addresses a critical limitation in current radio-interferometric pipelines, which typically achieve only a small fraction of peak hardware performance due to bottlenecks in memory and data transfer. astroCAMP introduces a unified suite of metrics covering scientific fidelity, computational performance, sustainability, and lifecycle economics, alongside standardised datasets for reproducible benchmarking across various computing platforms, including CPUs, GPUs, and emerging accelerators. Evaluation using established imaging algorithms demonstrates the potential for substantial improvements in energy efficiency and reductions in carbon emissions and total costs through co-design approaches. Specifically, the work highlights the interdependence of software and hardware optimisation, showing that improvements in software efficiency are crucial to unlock the full potential of energy-efficient hardware designs. The authors acknowledge that defining quantifiable fidelity metrics and acceptable tolerances for image quality remains a challenge for the SKA community. Addressing this limitation is crucial to enable principled optimisation and fully realise the benefits of co-design. Future work will focus on facilitating the definition of these thresholds and exploring heterogeneous CPU-FPGA designs to identify optimal operating points for SKA-scale imaging deployments. The release of datasets, benchmarking results, and a reproducibility kit aims to encourage wider adoption and collaboration within the SKA community. 👉 More information 🗞 astroCAMP: A Community Benchmark and Co-Design Framework for Sustainable SKA-Scale Radio Imaging 🧠 ArXiv: https://arxiv.org/abs/2512.13591 Tags: